인지과학연구학술지 04. 6. 9(사교)
4권2호 121-147
Milsark’s Generalization and Categorical Judgments
S.-Y. Kuroda
Linguistics Department, University of California, San Diego
sykuroda@ucsd edu
1. The objective of the paper
Milsark’s generalization is well known.
Milsark’s generalization “Properties may only be predicated of ‘strong’
NPs.” (Milsark 1977:16).
In this paper, I shall restrict myself mostly to the discussion of bare NPs. I
shall reformulate Milsark’s generalization with this restriction and in terms of
the individual-level (IL) and stage-level (SL) distinction introduced by
Carlson (1977). But I expect that the main claims to be made below can be
extended to indefinite NPs in general.1
Milsark’s generalization (restricted to bare NPs and rephrased in terms of
IL) “IL predicates may only be predicated of generic NPs.”
Counterpart. “SL predicates may be predicated of existential NPs.”
Milsark’s generalization has generally been assumed to be related to the
* I would like to thank Chris Barker, Gregory Carlson, John Moore and Susan
Fischer, who read early drafts of this paper and gave me valuable comments and
advice.
1 At least for some predicates, the distinction between SL/IL must be understood
relative to the sentences of which they are main predicates; in other words, the
distinction is strictly speaking about sentences, and about predicates only relative to
the sentences of which they are main predicates. In what follows I apply SL/IL to
sentences and predicates interchangeably.
Journal of Cognitive Science 4 : 121 - 147, 2003.
ⓒ2003 Institute for Cognitive Science, Seoul National University.
122
S.-Y. Kuroda
distinction between categorical and thetic judgments. The following
generalization in the theory of judgments must be responsible for this
perception:
The definiteness effect on categorical judgments2
(a) The subject of a CATEGORICAL JUDGMENT must be a “semantically
definite” (including “generic”) noun phrase.
(b) The subject of a CATEGORICAL JUDGMENT may not be an indefinite
“existential/cardinal” noun phrase.
The main objective of this paper is to challenge the validity of Milsark’s
generalization. According to the currently prevailing view, sentences with
bare NP subjects represent different types of propositions depending on
whether they are SL or IL; they are mapped onto different types of semantic
representations. In the former, subject NPs are existentially quantified and in
the latter they are bound by the generic operator. I will present evidence
against this view and argue that the distinction between SL and IL cuts
across the distinction between generic and existential construals of bare NP
subjects. I will maintain that sentences with bare NP subjects, whether SL or
IL, are subject to generic/universal or non-generic/existential construals
depending on the domain of interpretation.
2 Cf:
Kuroda (1965: 44). Ladusaw (1994) attempts to derive Milsark’s generalization
“from the assumptions involved in the ontology underlying the thetic categorical
distinction.” A step in his derivation corresponds to the definiteness effect cited
here. Two formulations are given by Ladusaw:
(i) The subject of a categorical judgment cannot be a nonspecific indefinite; its
reference is “presupposed.” (First version).
(ii) The subject of the basis of a categorical judgment must be an object, not a
description of an object. (Second Try)
(ii) may be compared with the following statement, where substance corresponds
to Ladusaw’s concept of object:
(iii) We might say that the ‘topic’ marker wa functions to indicate that the entity
referred to is cognitively apprehended as substance.” (Kuroda 1992: 42)
Milsark’s Generalization and Categorical Judgments
123
2. Preliminaries
It is well known that there are two sentence types in Japanese that
correspond to English declarative sentences: I call one the PLAIN SENTENCE and
the other the wa-MARKED SENTENCE. Examples of plain sentences are given in
(1) - (2) and examples of wa-marked ones are given in (3) - (5). The wamarked noun phrase is commonly called a topic. I will follow this
terminology, but I do not believe that the particle wa of the wa-marked
sentence can be accounted for as a topic marker in the sense understood in
discourse theory or information packaging. But I am not going to be
concerned with that issue in this paper. See Kuroda (1992: 15f, 70ff) and
Kuroda (2001).
(1) gengogaku no gakusei ga kinben de aru
linguistics GEN student NOM hard-working be
‘linguistics students are hard-working’
(2) inu ga neko o oikakete iru
dog NOM cat ACC chase be
‘dogs are chasing cats’
(3) gengogaku no gakusei wa kinben de aru
linguistics GEN student TOP hard-working is
‘linguistics students are hard-working’
(4) inu wa neko o oikakete iru
dog cat ACC chase is
‘dogs are chasing cats’
(5) neko wa inu ga oikakete iru
cat dog NOM chase is
‘cats, dogs are chasing’
In principle, any argument as well as any adjunct can be made a wa topic.
The syntactic subject is made a topic in (3) - (4), and the direct object in (5).
For the sake of this paper I assume the syntactic derivation of wa-marked
sentences from plain sentences by movement as given in (6). (3) - (5) are
derived by this movement as suggested in (7) - (9).
124
(6)
(7)
(8)
(9)
S.-Y. Kuroda
[CP ….[VP …DPi. ….]..] → [CPDPi-wa ..[VP ….ti ….]..]
[gengogaku no gakusei wa [t kinben de aru]]
[inu wa [ t neko o oikakete iru]]
[neko wa [inu ga t oikakete iru]]
Semantically, the wa-marked sentence cannot be truth-conditionally
distinguished from the plain sentence from which it is derived. To account
for the functional difference between wa-marked and plain sentences, I have
introduced and distinguished two semantic/cognitive objects, PROPOSITIONS
and JUDGMENTS. I follow the common assumption that sentences/clauses
represent PROPOSITIONS. In addition, I assume that sentences also express
JUDGMENTS in syntactic contexts where they make assertions.
I assume that wa-marked and plain sentences express different types of
judgments. A wa-marked sentence expresses a categorical judgment, that is,
a judgment in the form of a predication. A plain sentence expresses a
judgment of the nonpredicational form, a DESCRIPTION of an abstract or
concrete situation; the THETIC JUDGMENT is by definition a subtype of
description, a description of an event. In some embedded contexts, for
example in conditional clauses, sentences/clauses do not make assertions; in
such contexts only plain sentences can occur, and they are assumed to
represent propositions but not to express any judgments. For this
presentation, I have to assume that these distinctions are well known.3
In what follows, however, most of the time we can assume the common
semantic understanding according to which sentences are representations of
propositions. The distinction between propositions and judgments matters
only when I explicitly make reference to the concept of judgment. To sum
3
See Kuroda (1965, Chapter 2; 1972; 1992, Chapter 1). The concepts CATEGORICAL
and THETIC JUDGMENT originate in the Brentano-Marty tradition of linguistic
theory. They were applied to the account of the distinction between wa-marked and
plain sentences in Kuroda (1972), but this account, in terms of judgments, originates
in Kuroda (1965; chapter 2), where the distinction is made between PREDICATIONAL
JUDGMENT and NONPREDICATIONAL DESCRIPTION. A more detailed account of judgment and
sentence forms is given in Kuroda (1992: chapter 1). Kuroda (2001) extends the
concept of DESCRIPTION to IL sentences; the thetic judgment is a subtype of the
description, with an SL predicate.
JUDGMENT
Milsark’s Generalization and Categorical Judgments
125
up:
(10) a Plain sentences may express descriptive/thetic judgments and
represent propositions
b or simply represent propositions.
(11) Wa-marked sentences express categorical judgments and represent
propositions.
3. The interpretation of bare NP subjects of SL sentences
3.1. Let us consider the following SL sentence:
(12) a ringo ga
kago no
naka ni kireini narabete atta
apple NOM basket GEN in
at beautifully arrange was
b ‘apples were beautifully arranged in a basket’
The standard translation of (12) into formal logic is an existential proposition
given in (13):
(13) ∃x[APPLE(x)&ARRANGED-IN-A-BASKET(x)]
However, I would like to maintain that there are contexts in which (12) must
be understood as if it represents a universally quantified proposition.
Consider the following inference:
(14) (ano paathii de) ringo ga kago no naka ni kireini narabete atta; dakara
(that party at) apples NOM basket GEN in at beautifully arrange was;
therefore daremo ringo o tabe nakatta
nobody apple ACC eat neg-past
‘(at that party) apples were beautifully arranged in a basket; therefore
nobody ate apples’
Here, two sentences are connected by means of dakara ‘therefore’; this
conjoined sentence can give an impression that a natural inference was
made. The background of the reasoning is the common-sense understanding
that decent people refrain from disturbing what they perceive as a decorative
126
S.-Y. Kuroda
arrangement. But there must be another “logical” basis underlying this
natural reasoning: that the apples arranged in a basket are all the apples that
there were. (13), though true, is not sufficient to describe the logical function
of (12) in (14).
This point can be substantiated by observing the following subtle and
interesting contrast. Let us first translate the existential formula (13), the
standard logical rendition of (12), back into Japanese, literally by using the
verb of existence atta ‘was/were’. We get:
(15) a kago no naka ni kireini narabete aru ringo ga atta
basket GEN in at beautifully arrange be apple NOM was
b ‘apples that were beautifully arranged in a basket were/existed’
If we replace the first sentence (the premise) of the inference in (14) by (15),
we get:
(16) (ano party de) kago no naka ni kirei ni narabete aru ringo ga atta;
dakara daremo ringo o tabenakatta
‘(at the party) apples that were beautifully arranged in a basket
were/existed; therefore nobody ate apples’
The inferences (14) and (16) should be logically equivalent; we have simply
replaced one premise with another, which are supposedly logically
equivalent. However, there is a subtle, but remarkable difference between
(14) and (16). The same difference can be observed in English between (17)a
and (17)b:
(17) a Apples were arranged beautifully in a basket; so, nobody ate
apples.
b There were apples arranged beautifully in a basket; so nobody ate
apples.
(16), and I believe (17)b as well, lack the degree of naturalness of inference
we observe in (14) as well as (17)a. It is difficult, or impossible, to
understand ringo ‘apples’ in (16) and (17)b as universally quantified. It is
Milsark’s Generalization and Categorical Judgments
127
harder in (16) and (17)b than in (14) and (17)a to infer from the context that
there were no apples other than those in the basket. Let us put aside for the
moment why we get this difference, but the observed difference strengthens
our initial intuition about (14); that is, ringo in (14) can be understood as
universally quantified.
But it is definitely not the case that the subject ringo in (12) must be
interpreted as universally quantified in all contexts. It is possible to witness
an “inference” that contradicts such an interpretation.
(18) (ano paathii de) ringo ga
kago no naka ni kireini narabete atta;
daremo
(that party at) apples NOM basket GEN in at beautifully arrange was;
nobody
kago no naka no ringo o tabe nakatta
basket GEN in at apple ACC eat neg-past
‘(at the party) apples were beautifully arranged in a basket; nobody
ate apples arranged in the basket.’
By saying “nobody ate apples arranged in the basket” rather than “nobody
ate apples,” the speaker would naturally be taken as implying that people
might have eaten apples but those were apples that were not arranged in the
basket. To summarize, some uses of (12) require universal interpretation of
ringo ‘apples’ but other uses contradict such an interpretation. Nonetheless, I
would like to maintain that we have one and the same proposition
represented by (12), and the difference in question is the matter of domains
chosen for the interpretation of the proposition.
3.2. As a general description of the semantics of an SL sentence S with a
bare NP subject like (12), I propose the following formula:
(19) S = NP1 … Pred…e…,
where e is an event argument and NP1, the subject of the predicate.
Interpretation:
|NP1|S,w = |NP1|∩|e| where | | indicates the “universal” valuation
128
S.-Y. Kuroda
and | |S,w, the interpretation relative to S and a model/domain w
with respect to which the situation in question is described. |e|
denotes the set of entities involved in e.
Let us agree that (19) presupposes that w contains e; only in such a domain is
S interpretable. A domain w is called a g-DOMAIN for the proposition p
represented by an SL sentence S (or, simply, for S) if |NP1|S,w = |NP1|∩|e|.
Thus, a domain w is a g-domain for p, if and only if those “NP1’s” that
participate in e are exactly those that exist in w. Note that whether a domain
is a g-domain for p or not depends on the interpretation of the event
argument e. If a domain is not a g-domain, we call it a NON-g-DOMAIN.4
Let us consider sentence (12) in the context of the natural inference (14),
and examine how it is understood. The hearer understands that an inference
based on universal quantification is intended with respect to the subject,
ringo ‘apples’. Hence, the proposition p represented by (12) must be
interpreted with respect to a g-domain. The domain may be the entire
building where the party was held, or as small as the corner of the room
where the speaker and hearer were located at the time of the utterance; in any
event the domain is sufficiently small not to contain any apples other than
those arranged in the basket in question. In contrast, (12), given in the
context of discourse (18), must be understood as interpreted with respect to a
non-g-domain of (12).
The current common assumption is that the semantic representation (the
logical form) of an SL sentence with a bare NP subject is an existentially
quantified form like (13). This assumption does not account for the
understanding of (12) in (14). In order to account for this new situation, the
4
The definition |NP1|S,w = |NP1|∩|e| given in (19) is generally adequate only for the
case where Pred is a one-place predicate, as Chris Barker pointed out to me. Take, for
example, S = ‘fruits are beautifully arranged beside rotten apples’. Then, rotten apples
being fruits, |NP1|S,w as defined in (19) would contain rotten apples as well. In order
to avoid this difficulty, let |e|θ be the set of entities that participate in e with the theta
role θ of NP1 in S, and let |NP1|S,w = |NP1|∩|e|θ. According to this revision, no
domain is a g-domain for S = ‘fruits are beautifully arranged beside rotten apples’, as
desired. For our present purposes, we can restrict ourselves to one-place predicates
without losing generality.
Milsark’s Generalization and Categorical Judgments
129
common practice of formal semantics would require that sentence (12) be
assigned another logical representation, i. e., a universally quantified form.
One and the same sentence (12) would thus be assigned two logical forms;
the sentence would be considered as semantically ambiguous. Instead of
taking such a step, I propose that one and the same semantic interpretation
corresponds to (12), as indicated by (19). Then, the difference in the logical
function of (12) observed in (14) and (18) is a matter of different pragmatic
choices of domains with respect to which the sentence is construed. If the
domain is a g-domain, (12) can be an input to a universal type of inference. If
not, it cannot be, and it is subject to the usual existential construal given by
(13).
Semantic representations and model-theoretic interpretations are matters
of grammar, but the choice of the domain w with respect to which an
interpretation is executed is a matter of the creative aspect of language use.
Not only do choices constantly change during language use, but also they are
ambiguous, not well demarcated and hard to digitize. Logical laws are valid
only if w is kept constant through a discourse. This constraint is, practically,
hard to follow. In particular, g-domains for an event are generally very small,
and it is difficult/rare for a discourse to proceed with such a small discourse
domain being kept constant. Empirical evidence for g-domains may be hard
to come by.
Whether w is a g-domain or not, if (12) is truthfully asserted, (13) holds,
and vice versa. Thus, in this sense, (13) gives an adequate truth condition for
(12). But if w is a g-domain, (13) is an understatement and fails to monitor
possible arguments done with (12) in natural language. Insomuch as the
LOGICAL representation of a proposition represented by a sentence is assumed
to model reasoning in natural language performance, we have to assume
either that (12) is semantically ambiguous, representing two different
propositions, or that we cannot assign a unique logical representation to THE
proposition represented by (12). I am arguing for the second alternative and
against taking (13) as THE semantic representation of (12). Put in general
terms, my argument is that the logical representation of an SL sentence is a
matter of pragmatics and does not belong to the level of semantic
representation.
130
S.-Y. Kuroda
3.3. To recall, the contrast between the two inference forms, a natural one
(14) and an inadequate one, (16), was crucial for our judgment that a bare NP
subject may be construed as universally quantified. In order to account for
this contrast, let us assume that the default choice for the domain of
interpretation of an SL sentence with a bare NP subject is a g-domain.
This is a pragmatic default case assumption. Unless the pragmatic
situation contradicts it, one interprets an SL sentence S by choosing a
sufficiently small domain, so that it satisfies the condition for a g-domain.
We might try to accommodate our utterance to meet this condition. Imagine,
for example, that you are standing at a corner of a kitchen-dining area and
see apples arranged in a basket on a table but we also see apples put on plates
on the kitchen counter with other fruits. In this context, according to this
assumption, (21) would be preferred to (20):
(20) apples are arranged beautifully in a basket
(21) on the table, apples are arranged beautifully in a basket
The locative explicitly suggests that the intended domain of interpretation is
limited to a neighborhood of the table, excluding the counter area, a gdomain.5
According to our assumption, the unmarked choice for the domain of
interpretation for (12), the premise in (14), is a g-domain w of (12): There are
no apples in w other than those arranged in a basket as described by (12).
Given this understanding, (14) is a natural inference. Now, consider (15).
The subject of (15) is kago no naka ni kirei ni narabete aru ringo, ‘apples
arranged beautifully in a basket’, not ringo ‘apples’. What is a g-domain of
(15)? A domain W is a g-domain for (15) if and only if W does not contain
any apples arranged beautifully in a basket other than those arranged in the
5
This default case assumption is independent of the maximality condition associated
with the exhaustive listing implicature. Assume that some apples and oranges are
arranged in a basket. Given this context, (20) does not satisfy the maximality
condition in this sense, and yet satisfies the default case assumption, if there is no
apple that is not in the basket. Conversely, if there are some apples that are not in the
basket, (20) does not conform to the assumption, and yet it could satisfy the
maximality condition if apples but nothing else are in the basket.
Milsark’s Generalization and Categorical Judgments
131
basket referred to in (15); W may contain other apples not arranged in a
basket. Hence, with (15) as a premise, a universal inference about apples
arranged in a basket could be justified but not one about apples. As a
consequence, (16) does not sound like a natural inference, as (14) does.
(12) and (15) stand in a relation worthy of particular note. They are truthconditionally equivalent. Interpreted with respect to the same situation/event
e and with the same domain D, they are either true or false together.
Nonetheless, they are not logically equi-functional. A proper inference with
(12) as a premise ceases to SOUND proper if the premise is replaced by the
truth-conditionally equivalent (15). For, with (15), we are expected to
interpret the inference with respect to a g-domain of (15). But, in fact, for
(15), any domain is necessarily a g-domain, due to the special property of the
predicate exist: no domain can contain apples that are beautifully arranged in
a basket that do not exist. It follows that (15) is unprejudiced as to whether
there are apples other than those that are arranged in a basket, and (16)
cannot make a natural inference in any circumstances. We have thus
accounted for the functional difference between the truth-conditionally
equivalent (12) and (15).6
6
I argued that (i) cannot be assigned (ii), and for that matter any logical
representation, at the semantic level:
(i) a ringo ga kago no naka ni kireini narabete atta
apple NOM basket GEN in at beautifully arrange was
b ‘apples were beautifully arranged in a basket’
(ii)
∃x[APPLE(x)&ARRANGED-IN-A-BASKET(x)]
[= (12)]
[= (13)]
In contrast, one could argue that (iii) is effectively represented by (ii) at the
semantic level, due to the special property of the predicate exist:
(iii) a kago no naka ni kireini narabete aru ringo ga atta
basket GEN in at beautifully arrange be apple NOM was/existed
b ‘apples that were beautifully arranged in a basket were/existed’
[= 15]
For, the universally quantified proposition that a g-domain licenses for (iii) would
be:
(iv) ∀x[APPLE(x)&ARRANGED-IN-A-BASKET(x)⊃E(x)]
132
S.-Y. Kuroda
3.4. The following assertions follow directly from (19):
Assertion 1. (Downward invariance of the g-domain) A subdomain of a
g-domain for an SL proposition p with a bare NP subject is a g-domain for
p.7
Assertion 2. (Upward invariance of the non-g-domain) A domain that
contains a non-g-domain for an SL proposition p with a bare NP subject is a
non-g-domain for p.
Now I claim the following main assertion of this section to hold:
Assertion 3. An SL sentence S represents one and the same
through upward extension and downward restriction of domains.
PROPOSITION
Thus, with this assertion I contend that an SL sentence with a bare noun
phrase subject has a unique semantic representation, even though the logical
function of its subject is not invariant with changes of domains. One could
take this assertion either as a proposal for a way to use the term “roposition”
(a definition) or as a claim as to how to understand what “proposition” is (a
hypothesis). The reason that I put it as an assertion (thus a hypothesis and an
empirical claim) is that I assume we as theoreticians have some intuitive feel
representation for the semantic/pragmatic function of a sentence; (ii) suffices for
(iii) as a logical representation in any domain.
Now, consider the English there insertion sentence (v):
(v) there are apples arranged beautifully in a basket
On the one hand, (v) is commonly used as a translation of the Japanese sentence (i)
a as well as the logical formula (ii) into English. On the other hand, we observed
above a functional parallelism between the inference (17)b in English using (v) and
the inference (16) in Japanese using (iii) a. It is reasonable to assume, then, that (v)
is a surface transform of (iii) b, and not of (i)b.
Thus, for there insertion sentences it is legitimate to assume that existentially
quantified formulas like (ii) are assigned at the semantic level, even though they are
not for plain SL sentences like (i)b.
7 Presupposing that it is a subdomain where S is interpretable, that is, that it contains
the event e.
Milsark’s Generalization and Categorical Judgments
133
about “something” that we would wish to call a proposition and Assertion 3
is about this something. It claims that if the subject of an SL sentence is a
bare noun phrase, its apparent function changes between g-domains and nong-domains, but the sentence represents one and the same “proposition”; it is
not SEMANTICALLY AMBIGUOUS. The speaker and hearer are to be warned that for
a proper application of logic, they must keep track of the changes of domains
of interpretation properly during a discourse, rather than of the changes of
propositions represented by one and the same sentence.
So far the Japanese sentences we have dealt with are all plain sentences.
Let us at this point consider wa-marked sentences. Consider (22):
(22) ringo wa kago no naka ni kireini narabete atta.
apple top basket gen in at beautifully arrange were
‘apples were beautifully arranged in a basket’
Let us substitute (22) for (12) in (14):
(23) a (ano paathii de) ringo wa kago no naka ni kirei ni narabete atta;
dakara daremo ringo o tabenakatta
b ‘(at that party) [the] apples were arranged beautifully in a basket;
so, nobody ate apples’
I translated ringo wa as ‘[the] apples’, with the definite article, but as there is
no syntactic distinction between definite and indefinite in Japanese, it is a
moot point to assume that the wa phrase in (23) here must be categorially
taken as definite.8 What interests us, and what is not moot, is that the wamarking carries the implication that the apples in the basket were all the
apples there were. Thus, (23) sounds as natural as, indeed perhaps more
8
Ringo wa in (23)a could refer to the apples that have already been mentioned in the
course of the discourse, or the specific apples implicitly understood in the context, but
it does not have to and in that case, the different effect of (23) attributable to wa
seems to be that somehow or other the presence of apples are expected at parties or a
particular party of the topic in the discourse. According to native speaker consultants,
under such circumstances, the use of the definite NP the apples is proper in English,
even though it is not anaphoric.
134
S.-Y. Kuroda
natural than, (14) as a deduction based on the common-sense presupposition
that decent people do not disturb a decorative arrangement.
Compare (18) with (24) below. I replaced ga in (18) by wa.
(24) (ano paathii de) ringo wa kago no naka ni kirei ni narabete atta;
daremo kago no naka no ringo o tabe nakatta
‘(at that party) (the) apples were beautifully arranged in a basket;
nobody ate apples in the basket.’
(18) invites the inference that there were other apples than those that were
arranged in the basket. In contrast, (24) does not naturally invite such an
inference. Rather, the wa after the first ringo suggests that the apples in the
basket were all the apples that there were. With this understanding, the
inference would go through, though one might at least sense stylistic
awkwardness: the direct object of the second sentence sounds redundantly
modified by kago no naka no ‘in the basket’.
These observations show that the wa-marked sentence (22) must be
interpreted in a g-domain. I take (23) - (24) as evidence for the following
assertion.
Assertion 4. A wa-marked SL sentence with a bare noun phrase subject is
interpreted in a g-domain of the proposition represented by its plain sentence
source.
4. The interpretation of bare NP subjects of IL sentences
We now move to IL sentences. Consider the following discourse (25):
(25) a France-go wa
iroirona kuni no
hito ga hanasite iru
France-language various country GEN person NOM speak are
‘persons of various countries speak French’
b (Motiron,) France-zin
ga
France-go
o hanasu
of course France-people NOM
France-language
ACC speak
‘(Of course) The French speak French’
c (sorekara) Morocco-zin ga
France-go
o hanasu
Milsark’s Generalization and Categorical Judgments
135
and
Morocco-people NOM France-language ACC speak
‘(and) Moroccans speak French’
d Canada-zin
ga
France-go
o hanasu.
Canada-people NOM France-language
ACC speak
‘Canadians speak French’
The intended meaning of (25)d is not “generic/universal,” but only
“existential.” The possibility of this non-generic/non-universal construal of a
plain IL sentence contrasts sharply with the impossibility of such a construal
for a wa-marked IL sentence. (28) is approximated only by (29); it’s
construal cannot be “non-generic/non-universal” as (26) can. Thus, the plain
sentence (26) can be used as an “existential” statement about Canadians in
this real world, as in discourse (25), while (28) in contrast cannot be
truthfully uttered about Canadians in general, given the way the real world is.
(28) can of course be truthfully uttered, but only in special circumstances
where all Canadians present happen generally to be French-speaking.
(26) Canada-zin ga France-go o hanasu
(27) Some Canadians speak French
(28) Canada-zin wa France-go o hanasu
(29) All Canadians speak French
[= (25d)]
[contrual of (26) in (25)]
Keep in mind that the contrast between the wa-marked and the plain
sentences is not between generic/universal and non-generic/non-universal
interpretations; the plain sentence form, like (25)b-d, does not exclude the
possibility of a generic/universal interpretation. Rather, the syntactic and
pragmatic context determines the interpretation. Take, for example, the plain
sentence form (25)b. Even in the context of discourse (25), this sentence can
be taken as intended to be a generic/universal statement; it could in fact be
continued as follows:
(30) France-zin
ga
France-go
o hanasu; zissai,
France-zin wa
France -people NOM France-language ACC speak; in fact, France-people France-go
sika hanasanai
136
S.-Y. Kuroda
France-language except speak-neg
‘The French speak French; in fact, the French speak only French.’
Lest one should wonder if the predicate hanasu ‘speak’ in generic
interpretation might be taken as a type transfer from an SL predicate ‘be
speaking’ and suggest that what we have observed about this IL predicate is
due to this type transfer, let me give examples with predicative nominals,
intrinsically IP predicates:
(31) (yo-no-naka mo kawatta mono da)
(world
also
changed thing is)
‘(the fact is that the world has indeed changed)’
Nihon-zin ga major league no sensyu de Mooko-zin ga sumoo tori da
Japanese NOM GEN player and Mongolian NOM sumo wrestler is
‘Japanese are major league players and Mongolians are sumo
wrestlers’
(32) (yo no naka mo kawatta mono da)
major no sensyu ga nihon-zin de sumoo tori ga mooko-zin da
‘(the fact is that the world has indeed changed)
major league players are Japanese and sumo wrestlers are
Mongolians’
The intended interpretation of (31) is not of course that all, or even many,
major league players are Japanese, nor that all, or even most sumo wrestlers
are Mongolian.
To recapitulate, whether we have a generic/universal construal or nongeneric/non-universal construal of a bare NP subject depends on context, that
is, on domains of intended interpretation. Thus, what we have with IL
sentences is a situation much like we have had above with SL sentences.9
9
Cohen (2001) also notes “generics [that] receive what appears to be an existential
interpretation.” His existential generics, however, are accompanied by emphasis or
focus. (26), given in the context in (25), may also be taken as implicitly focused, but
(31)-(32) are plain statements without focus and show that existential construals of IL
sentences are not limited to those with emphasis or focus.
Milsark’s Generalization and Categorical Judgments
137
Given these observations, I maintain that as in the case with SL sentences,
whether a generic/universal or “non-generic/existential” interpretation is
intended for the bare NP subject of an IL sentence is a matter of domains
with respect to which the sentence is evaluated, and not the matter of
whether the semantic representation of the sentence has a categorially
distinct generic or existential NP as the subject. This point is obscured as
long as our attention is turned only to generic sentences of the type which
allow usual generic interpretation due to the way the real world actually is,
such as (25)b above. (25)b can be interpreted as generic/universal whatever
the domain of interpretation happens to be. Examples like (26) and (31) are
crucial.
We might say that an IL sentence with a bare NP is construed as
generic/universal if the intended domain of interpretation is a g-domain, and
as non-generic/non-universal (existential) if not. However, we cannot define
g-and non-g-domains for IL sentences similarly to way we did for SL
sentences. For one thing, it is not obvious that the introduction of an
analogue of (19) for IL sentences, with a “state of affairs” argument in place
of an event argument, is a viable solution. Furthermore, as is well known, a
generic reading is in many ways different from a universal reading and defies
any simplistic account in terms of set-theoretically defined extensions. Nor is
it clear exactly how the non-generic/non-universal construal of a bare NP
subject of an IL sentence should be understood. It is certainly an existential
construal of some sort, but there is something generic about it, too; it might
be called an existential generic. Mere existence does not suffice to license
existential generics. For example, even though there are French speaking
Americans, it would not be proper to say “Americans speak French” in the
context of (25). I shall not attempt to interpret/define g-domains for IL
sentences set/model theoretically. Instead, I adopt an axiomatic approach; by
reversing the direction of conceptualization, I propose to understand gdomains for IL sentences as those that license generic/universal construals
and non-g-domains as those that do not.
I maintain, then, on the strength of Assertions 1-4, which followed from
(19), that the following analogues of Assertions 1-4 hold as axioms:
Analogue of Assertion 1. (Downward invariance of the g-domain) A
138
S.-Y. Kuroda
subdomain of a g-domain for an IL proposition p with a bare NP subject is a
g-domain for p.10
Analogue of Assertion 2. (Upward invariance of the non-g-domain) A
domain that contains a non-g-domain for an IL proposition p with a bare NP
subject is a non-g-domain for p.
Analogue of Assertion 3. An IL sentence S represents one and the same
PROPOSITION through upward extension and downward restriction of domains.
Analogue of Assertion 4. A wa-marked IL sentence with a bare noun phrase
subject is interpreted in a g-domain of the proposition represented by its
source plain sentence.
These are plausible assertions, but since we lack grounding of the basic
concepts on the standard set/model theoretic framework, they must be taken,
and ultimately justified, as hypotheses for an empirical theory about g- and
non-g-domains, or, equivalently, about generic/universal and nongeneric/existential construal.
5. The synthesis
In the preceding two sections I have presented evidence for the claim that
sentences with bare NP subjects, whether SL or IL, are subject to
generic/universal or non-generic/non-universal, existential construals
depending on whether the domain of interpretation is a g-domain or not. I
further maintain that parallel sets of assertions hold about these construals
and domains for SL and IL sentences. A standard response to the situation
we face might be to assume that both SL and IL sentences are semantically
ambiguous, mapped onto two distinct semantic representations. Formally,
such a standard semantic approach would adequately describe the observed
distinctions. However, I would propose a different, pragmatic approach. I am
not at present in a position to offer empirical evidence to choose one or the
10
I am not certain what presupposition, if any, is required for a subdomain where IL
proposition p remains interpretable; see note 7 for the condition that subdomains
contain the event e for the SL case. This Analogue must be understood as an axiom in
conformity with which the relevant sense of “subdomain” must be determined.
Milsark’s Generalization and Categorical Judgments
139
other approach on formal grounds. My argument is a plausibility argument
speculating on performance.
Thus, as stated in Assertion 3 and its analogue above, I propose that a
sentence with a bare NP subject represents one and the same proposition, but
that its construal depends on where it is to be interpreted. I would attribute to
the speaker and hearer more secure knowledge about sentences and
propositions than merely how to interpret them, which depends on the choice
of the domain for interpretation. Misunderstanding, if it takes place, is then
due not to the failure of understanding one and the same sentence or one and
the same proposition (meaning) represented by a sentence by the speaker and
hearer, but to the failure to agree on the pragmatic choice of one and the
same domain for the interpretation of the communicated proposition.
In view of the parallelism we observed above between SL and IL
sentences with bare noun phrase subjects, I conclude:
Thesis 1. Sentences with bare NP subjects represent the same type of
propositions, regardless of whether they are SL or IL.
Let us call bare NP subjects as well as sentences with bare NP subjects
GENERIC/EXISTENTIAL. Generic/existential sentences satisfy the same set of
axioms formulated in Assertions 1-4 and their Analogues whether they are
SL or IL.
There is nonetheless a difference between the SL and the IL cases. For SL
sentences, we can adequately account for both generic/universal and nongeneric/existential interpretation and for g- and non-g-domains in terms of
the standard set/model theoretic approach to semantics. For IL sentences, we
do not have a satisfactory account, either for generic/universal or for nongeneric/existential interpretation; nor can we ground the concept of g-domain
on a set-theoretic basis. This means that we cannot present a unified
extensionalist account for the attempted generalization over the SL and IL
distinction formulated in Thesis 1.
Be that as it may, I still have to formulate a generalization for the unified
understanding of generic/universal and non-generic/existential interpretations
for SL and IL sentences. Let us leave aside negative sentences, as we have in
effect been doing in this presentation, since negation involves its own
140
S.-Y. Kuroda
complications. Then, the valid generalization is: a sentence with a bare NP
subject can be construed as representing a universal affirmative proposition
in the sense of classical syllogism (that is, the A form in the traditional
notation) when intended to be interpreted in a g-domain, and a particular
affirmative (that is, the I form) when intended to be interpreted in a non-g
domain. A generic/existential proposition can be an input to an A-type
inference if interpreted in a g-domain, and to an E-type inference when
interpreted in a non-g-domain. Let me conclude:
Assertion 5. A generic/existential (affirmative) sentence is interpreted as the
A form (universal) and the I form (particular) of a syllogism in a g-domain
and non-g-domain, respectively, and respective syllogistic inference rules
may apply.
Let us note that whatever the right LOGICAL representations of
generic/universal and non-generic/existential interpretations might be for IL
sentences, they cannot in general be taken as their semantic representations;
the same point has been claimed for the logical representations of SL
sentences in the form of existentially and universally quantified formulas of
predicate calculus in section 3.2 above. Assertion 5 must be understood as
entailing the following general claim:
Thesis 2. Logical representations of sentences are not semantic
representations; they are pragmatically determined relative to domains of
interpretation.
6. Milsark’s generalization(s)
With the preceding preparations, we now proceed to discuss Milsark’s
generalization. I repeat it here in two forms, as before:
Milsark’s generalization: Properties may only be predicated of “strong”
NPs. (Milsark 1977:16).
Milsark’s generalization (restricted to bare NPs and rephrased in terms of IL)
IL predicates may only be predicated of generic NPs.
Milsark’s Generalization and Categorical Judgments
141
Let us remind ourselves that Milsark’s original definition of the
“strong/weak” distinction is syntactic in character, in fact, dependent on
English syntax: NPs are weak if they are permitted in post-copular position
in there-insertion existential sentences; otherwise, they are strong (Milsark
1977: 8). The distinction, however, has been extended to other languages in
later literature, on semantic grounds. For our present consideration of bare
NPs, the distinction amounts to that between generic and existential. In any
case, the original Milsark generalization concerns selectional restrictions that
hold between predicate types and NP types. As far as bare NPs are
concerned, I have maintained that the categorial selection between generic
and existential is illusory, and hence so is that subpart of Milsark’s
generalization that relates to this distinction. I put this conclusion in the
following thesis.
Thesis 3. Milsark’s generalization (restricted form) is an epiphenomenon.
The argument for this thesis prima facie cannot be directly extended to
cardinality NPs such as two apples, many apples etc. We need to clarify what
we want to understand by “generic/universal cardinals.” Be that as it may, I
believe that Thesis 3 can be generalized to the original form of Milsark’s
generalization in a full form, but in this paper, I put cardinality NPs aside and
leave the matter only as a conjecture.11
There is, however, a distinct difference between SL and IL sentences: SL
11
Besides, as far as Japanese is concerned, complications involved in the syntax and
semantics of numeral or cardinality expressions must be worked out before any
substantive arguments can be made on them. For example, the following two forms
of slightly different syntax seem to show varied acceptability.
(Yononaka mo kawatta mono da)
(the world has changed)
(i) Nihon-zin ga san-nin (mo) yuumeina major no sensyu da.
Japanese NOM three-CLS (also) famous major GEN player be
‘(As many as) three Japanese are famous major league players’
(ii) ?*San-nin (no) Nihonzin ga yuumeina major no sensyu da
three-CLS
GEN
Japanese NOM FAMOUS major GEN player be
‘three Japanese are famous major league players’
142
S.-Y. Kuroda
sentences allow referential construal of their indefinite subject but not IL
sentences.
Thesis 4. (The residue of Milsark’s generalization) Indefinite NPs can be
‘referential’ in SL sentences but not in IL sentences.
Observe the contrast between (33) and (34) and between (35) and (36):
(33) Look there, (many) apples are arranged beautifully in a basket!
(34) *Look there, (many) apples are beautiful!
(35) Look there, (many) Canadians are speaking French
(36) *Look there, (many) Canadians speak French
Under Milsark’s generalization, this restriction might be understood as a
corollary: the referential reading is dependent on and derived from the
existential reading. Milsark’s generalization excluded existential readings
from IL sentences, hence only SL sentences would be able to have referential
readings.
This account, however, cannot be kept intact in our new perspective: we
allow non-generic/existential readings for IL sentences as well, and hence the
existential/non-existential distinction cannot be employed to single out SL
sentences. Referentiability does not cut across the SL/IL distinction, that is,
the event/property distinction. Referentialibility seems to be a fundamental
characteristic feature of event-hood. The referential-nonreferential distinction
amounts, it appears, to restating the definition of the IL/SL distinction; this is
the significance of Thesis 4.
Let us try to derive this generalization directly from the SL/IL distinction.
Let us recall the form of the representation of an SL proposition given in
(19), which I repeat here as (37):
(37) Representation of an SL proposition p:
NP1 … Pred…e…,
where e is an event argument and NP is the subject of the predicate.
If we take e as a variable, and unselectively bind (37) we get an existential
Milsark’s Generalization and Categorical Judgments
143
construal:
(38) Existential reading: ∃ (e, x)[NP1(x) … Pred…e…]
If we take e as a constant and do not bind e, we assume that we get a
referential construal:
(39) Referential reading: |NP1|S,w = |NP1|∩|e|, where | | indicates the
“universal” valuation and | |S,w the interpretation relative to S and a
model/world w with respect to which the situation in question is
described. |e| denotes the set of entities involved in e.
The referential interpretation of NP1 falls out from the referential construal of
the event argument e. We assume that this is the only way indefinite NPs get
referential readings. That is, indefinite REFERENTIAL NPs are dependent on events.
So far, we have discussed Milsark’s generalization within the bounds of
the theory of propositions, where its original form belongs. Let us now move
to the discussion of the perceived connection between Milsark’s
generalization and the theory of judgments. In Japanese, the subject of a
categorical judgment (i.e., the wa-phrase of a wa-marked sentence) must be a
semantically definite noun phrase, assuming that generic bare NPs are
semantically definite. We seem to be able to state a generalization much like
Milsark’s: The subject of a categorical judgment must be strong.
But in our approach, this formulation is problematic; for, the strong/weak
distinction as originally understood concerns a categorial distinction. But we
do not distinguish generic (i.e., strong) bare NPs from existential (i.e., weak)
bare NPs as a different category. We have only one kind of bare NPs; i.e.,
bare NPs are generic/existential NPs, though they may function
pragmatically in distinct ways, as they could participate in the
generic/universal type of inference or the non-generic/existential type of
inference, depending on pragmatic choices of domains. But according to
Assertion 4 in section 3.4 and the Analog of Assertion 4 in section 4, a wamarked bare NP, that is, a bare NP as the subject of a categorical judgment,
must be interpreted in such a way that the wa-marked subject can be an input
to a generic/universal inference. This condition must be characterized in
144
S.-Y. Kuroda
terms of domains in our approach. To repeat, compare (40) and (41):
(40) ringo wa kago no naka ni kirei ni narabete atta
‘(the) apples are arranged beautifully in a basket’
‘there are apples arranged beautifully in a basket’
(41) ringo ga kago no naka ni kirei ni narabete atta
‘(the) apples are arranged beautifully in a basket’
[=(22)]
[=(12)]
The plain sentence (41) could be interpreted in a g-domain; in fact, I have
even suggested that that is the default choice. Be that as it may, (41) can be
interpreted in a non-g-domain as well. In contrast, the wa-marking of the
subject in (40) excludes this possibility; the wa-marking signals that the
interpretation is intended in a g-domain. That means that the speaker must
commit himself/herself to such a possibility of the subject term being an
input to a generic/universal (A) type of inference. Likewise, compare (42)
and (43):
(42) Canada-zin wa France-go o hanasu
‘(the) Canadians speak French’
(43) Canada-zin ga France-go o hanasu
‘Canadians speak French’
[= (28)]
[= (25)d]
The real world being what it is, the wa-marked sentence (42) can be accepted
as a true statement only if the context is such that the generic statement “(the)
Canadians present in the context speak French” is appropriate; that is, the
context must be a g-domain. (43), on the other hand, can be taken as a true
statement about the real world itself, which is not a g-domain. In contrast,
(44) is an adequate statement in any context as long as the discourse
concerns the real world:
(44) France-zin wa France-go o hanasu
‘The French speak French’
We thus have a Milsark type of constraint, restricted to bare NPs, transferred
to the theory of judgment and formulated in pragmatic terms:
Milsark’s Generalization and Categorical Judgments
145
Thesis 5. (An analogue of Milsark’s generalization) A categorical
judgment must be interpreted in a g-domain of its subject.
This thesis is related to the definiteness effect on categorical judgments
mentioned earlier:
The definiteness effect on categorical judgments
(a) The subject of a CATEGORICAL JUDGMENT must be a “semantically
definite” (including “generic”) noun phrase.
(b) The subject of a CATEGORICAL JUDGMENT may not be (construed as) an
indefinite “existential/cardinal” noun phrase.
“Definite” and “indefinite”are mutually exclusive and hence (b) follows
from (a) as a corollary. The definiteness effect applies to non-bare definite
noun phrases such as sono ringo ‘that apple(s)’ and proper names, but Thesis
5 appears not to, since g-domain has so far been defined only for bare noun
phrases. But Thesis 5 can be understood to cover definite noun phrases as
well by extending the concept of g-domain to definite noun phrases: indeed
in conformity with the definition of definiteness, any domain where a
definite noun phrase can properly be used may be understood as a g-domain
for it. Then, Thesis 5 can be taken as equivalent to the formulation of the
definiteness effect given in (a) above.
The advantage of Thesis 5 over the definiteness effect is that it does not
refer to the definite/indefinite distinction, which is not syntactically marked
in Japanese. We do not have to decide whether ringo ‘apples’ in (40), or, for
that matter, France-zin in (44), either, is definite or not.
7. Summary and conclusion
In the view expressed here, sentences with bare NP subjects (or,
equivalently, the propositions represented by them) are of one kind,
generic/existential, representing the same type of propositions, whether they
are SL or IL (Thesis 1); the different construals of the subjects, either as
generic or existential, are a matter of pragmatic choices of domains. More
generally, logical representations of sentences are not semantic
146
S.-Y. Kuroda
representations (Thesis 2).
The original form of Milsark’s generalization is about an epiphenomenon (Thesis 3) and needs no account (or, is accounted for by the
recognition that it represents an epiphenomenon). However, we recognize a
characteristic difference between SL and IL sentences with respect to the
referential construal of indefinite NPs, the Residue of Milsark’s
generalization (Thesis 4). The original epiphenomenon is couched in the
distinction between SL/IL predicates, but when this epiphenomenon
disappears as an epiphenomenon, what is left, the residue, is what couched it.
Milsark’s generalization originally belongs to the theory of propositions,
but it seems to have commonly been perceived that it somehow relates to the
theory of judgments; it might even have been expected that the theory of
judgments accounts for Milsark’s generalization. But Milsark’s
generalization, on closer examination, turns out to be an epiphenomenon;
thus, nothing is left in the theory of propositions for which the theory of
judgments could render service to it by deriving it. Nonetheless, we
recognize a phenomenon that belongs to the theory of judgments which
could be thought of as responsible for the perceived relation of Milsark’s
generalization to the theory of judgments. In our approach, this phenomenon
can be identified as a matter of pragmatic choices of the domains for the
interpretation of categorical judgments, An analogue of Milsark’s
generalization (Thesis 5). We have thus a generalization of a character quite
different from Milsark’s original, but I have retained Milsark’s name for the
sake of the perceived relation to its namesake.12
12
The original talk I gave at the SALT meeting was entitled “Categorical and thetic
judgments, Milsark’s generalization and the definiteness effect.” In a later part of the
talk, I tried to elaborate on the contrast between the two types of “existential”
sentences, illustrated by (12) and (15), respectively, and the implication of that
contrast for how thetic judgments are expressed. I called a sentence of the latter type
the existential transform of the corresponding one of the former type. Syntactic
ecology, so to speak, of a particular language could influence the profiles of these
truth-conditionally equivalent but logically not equi-functional sentence types in that
language. I claim that in English existential transforms are represented by thereinsertion sentences at the surface level. See note 6 above. Now, the thetic judgment is
often (but should not be exclusively) associated with “existential” sentences, and for
Milsark’s Generalization and Categorical Judgments
147
References
Carlson, Gregory N. (1977). Reference to Kinds in English. U. MA Dissertation.
[Published from New York: Garland Press, 1980.]
Cohen, Ariel. (2001). Existential generics. Ben-Gurion University of the Negev, ms.
Kuroda, S.-Y. (1965). Generative Grammatical Studies in the Japanese Language,
Dissertation, MIT. Facsimile edition, New York: Garland Press, 1979.
Kuroda, S.-Y. (1972). “The categorical and the thetic judgments; evidence from
Japanese syntax,” Foundations of Language 9.2:153-185.
Kuroda, S.-Y. (1992). Japanese Syntax and Semantics: Collected Papers. Dordrecht:
Kluwer Academic. In particular, Chapter 1. Judgment forms and sentence
forms.
Kuroda, S.-Y. (2001). Focusing on the matter of topic: a study of wa and ga in
Japanese. Talk presented at Charles University, Prague.
Ladusaw, William A (1994). “Thetic and categorical, stage and individual, weak and
strong.” In M. Harvey and L. Santelmann eds. Proceedings of SALT IV, Ithaca,
NY, 220-229.
Milsark, G. L. (1977). Toward an explanation of certain peculiarities of the
existential construction in English. Linguistic analysis 3.1-30.
English, in particular, with there-insertion sentences, therefore with existential
transforms, in the sense introduced above. Due to this association, the thetic judgment
appears also often associated with the “definiteness,” or rather, indefiniteness effect.
In contrast, in Japanese, the existential transform should count rather as a marked
form, and as a consequence, the plain form is a normal expression for a thetic
judgment; Japanese thus is clear of the perceived (in)definiteness effect of thetic
judgments.
Paying attention to the functional contrast between sentences of the plain form and
their existential transforms was instrumental for initiating the approach developed
above. It would also be beneficial for a proper understanding of the thetic judgment
beyond the bounds of truth conditional equivalence. But this topic is tangential to the
main theme of this paper and I leave it for future work.